High Level Design Discussion

[ShahN005]

April 30, 2017

people's joys and emotions over the 10 past years

• can we find the correlation between an emotion and a company

prototype #1: find and publish (paid) the emotions towards all the companies on the globe

• read a tweet, company name, and find/read the emotion
• circulate through all the countries, for all companies, all occurrences.
• we read a tweet, does it have the company name, emotion distribution, all the fields, record it with a timestamp. that's the info
• question: is the tweet that is written about walmart a happy or angry tweet?
  • when people were continually joy/angry what was the reason?
• change the corpuse, and teach it how to react to that.

-- Askari's confidence level is around 70%, please re-approve.

prototype #2: sentiment analysis

• Askari and Nazia Sentiment Analysis Company

Stage 3: contextualize the tweets. not looking at the person. expanded on prototype #1

• companies sentiments on tweets, what are companies tweeting
• can keep adding daily
• doesn't have to be only for organizations. celebrities, politicians, etc

[ShahN005]

May 07, 2017

• is it possible to probe the gender of a tweet using the twitter name? million dollar question --> YES

  • can you teach watson to distinguish between male/female name
  • part of metadata: eg. e ending names in french are more likely female

• for protype #1: what to filter in the twitter cloud

  • eg. keyword search
  • browser search vs. human search
  • we would only do the search for the company that would pay us for it, use that as a template: keep changing the filters
  • do you want the data for overall or do you want it for a certain time period
    • validation testing: pre-pepsi ad vs. post ad (sentiment analysis over the # of days; one day before the ad and one week after, each day)
      • PHP Insight Engine:
        1. what do people feel about this company
        2. how people felt about pepsi one day before the ad
        3. how people felt about pepsi the day the ad aired, the second day and each day following up to a week up until today
  • filter template
  • creating a table and each row will contain a) tweet ID b)timestamp c)organization
  • create another table containing the a)tweet ID b)sentiment c)ranking
  • eg. involving both tables using 1 million tweets: group them by sentiment
  • then filter the data based on timestamp, piece the data to see what kinda curves it takes

• use case: maybe there are outside influences that companies/people haven't thought of that influence the sentiment about them

• eg. google trend

[ShahN005]

Process Outlines

May 7, 2017